Deep Dive: Creating an AI music video with MiniMax H3 workflow
TheDerminator1337 · reddit · 2026-08-31
The author details the workflow for creating a 1girl music video using the MiniMax H3 model. The process involves using Krea 2 for face, body, and background generation, then applying MiniMax image editing to create clothing for reference. Each clip uses three references (head, body with clothes, background) to ensure consistency. The setup uses 2MP resolution with sparse attention and 20 sampling steps, taking 5-6 hours per clip on a 5090 GPU. Custom nodes were used for granular per-clip control. The author notes that 2MP resolution aids facial detail and that face refiner tools conflict with camera cuts.
More from Multimodal
- Testing REFMOD: Generating consistent characters without reference images — chaindrop · 2026-08-31
- Dan Brickley Discusses Feasibility of 3D Projection for WebXR — danbri · 2026-08-31
- PromptNook: Open-Source Local Prompt and Model Catalog for Image-Generation Workflows — bayf0resT · 2026-08-31
- Same Prompt Test: ChatGPT Beats Gemini in Y2K Anime Style Generation — reayen · 2026-08-31
- Museum displays AI-generated art with 'No AI Used' disclaimer — Sauers_ · 2026-08-31
- Porting MiniMax H3 FastVideo LoRA to ComfyUI: 3x speedup achieved — Sad_Berry_4621 · 2026-08-31